RU2754476C1 - Method for vibrational diagnostics of technical condition of gas turbine engines in resource-conserving modes using invariant theory - Google Patents

Abstract

FIELD: testing.SUBSTANCE: present invention relates to methods for vibrational diagnostics of intermittent mechanisms, in particular, to a method for vibrational diagnostics of gas turbine engines (GTE). The substance of the invention consists in identifying certain characteristics of the object, which remain unchanged during normal operation of the object and change with occurrence of defects. The characteristics (invariants) are then used as direct or indirect diagnostic indicators. Realised in the method is a possibility of applying the apparatus of cumulants of higher-order spectra to obtain a system of diagnostic indicators and invariants suitable for identifying the early stages of development of damage to the bearing supports of a GTE, including at rotor rates corresponding to the cold or technological engine motoring modes.EFFECT: method makes it possible to increase the accuracy, certainty and efficiency of diagnostics of gas turbine engines in resource-conserving modes of operation.1 cl, 9 dwg

Description

The present invention relates to methods for vibration diagnostics of intermittent mechanisms, namely, to a method for vibration diagnostics of a gas turbine engine (GTE). In practice, the proposed method can be used to monitor the vibration state of gas turbine engines of sea and river transport, ships and vessels of the Navy, and can also be used for vibration diagnostics of engines used in the electric power industry and for gas transportation.

At present, methods for determining the state of various technical objects during vibration diagnostics are quite widespread, including obtaining vibrodiagnostic parameters in the form of a vibration signal (movement, speed, acceleration, etc. of the object under study) in the time domain, and its subsequent processing. These methods are based on the fact that during the operation of the diagnostic object, dynamic processes cause vibrations of both moving and stationary parts of the mechanism, which are transmitted to its body. To diagnose the mechanism, a vibration signal is measured and, based on its analysis, a conclusion is made about the state of the object. The main disadvantages of the methods [1-3, 5, 6], which in technical essence are most suitable for the proposed invention, are:

- carrying out diagnostics of the object in operating modes;

- the possibility of fixing already existing defects of the object of diagnostics, when modern operating conditions require the identification of defects at the stage of inception;

- control of the general level of vibration [3, 5, 6], in the rotor frequency band, or measurement and control of vibrations at the frequencies of the first harmonics of the rotors, which allows solving the problems of emergency protection of gas turbine engines, but does not allow solving the problems of effective diagnostics of malfunctions of rotor bearings at an early stage even using various methods of trend and probabilistic analysis of the overall vibration level.

There is a known method of vibrodiagnostics of the technical state of piston machines [7], which does not coincide in the object of research, but is very similar in the method of processing the diagnostic results. The disadvantages of the proposed method are:

- the method is considered in relation to piston machines (the difference not only in purpose, but also in the design of the technical devices under consideration cannot be guaranteed to give an answer in the correctness of the application of this method for diagnosing a gas turbine engine, in the reliability of the results obtained and requires additional research);

- the object is diagnosed in operating modes;

- the authors analyze the level of spectral invariants of the amplitude-frequency spectrum of vibration of the body of the diagnosed piston machine, taking into account the dependence of the level of spectral invariants on the technical state of its parts (in particular, valves of a piston compressor, crosshead, parts of the cylinder-piston group of a piston machine, etc.). The work does not pay attention to the establishment and study of changes in correlations between the measurement channels in the vibration diagnostics system, which tend to change over time with respect to the development of various defects and can improve the quality of diagnostic studies.

The prototype of the proposed method is a method for vibration diagnostics of a gas turbine engine in the modes of cold scrolling and rundown of rotors, presented in [4]. In their work, the authors propose the following algorithm of actions: recording the vibrations of a gas turbine engine with the required discretization parameters in cold cranking (CP) modes, obtaining a cascade spectrum of vibration signals with the further identification of the first harmonics of the rotors, according to which the bearing components are calculated and the results obtained are compared with a certain reference value according to to which a diagnosis of the technical condition of the object is developed. Disadvantages of the prototype method:

- the method is considered in relation to aviation GTE. Aviation GTEs, unlike GTEs for sea and river transport, as well as GTEs used in the electric power industry and for gas transportation, have a number of not only design features: large weight and size characteristics, the presence of a propeller (generator shaft, gas blower) in the turbine design, active jet execution of the flow path, operating parameters, etc.), as well as distinctive features in operating conditions: high air humidity and the presence of seawater droplets in the air, which lead to uneven scale deposition along the length of the flow path (for gas turbine engines of sea and river transport); the presence of dust in the atmosphere, which negatively affects many units of the gas turbine engine (for gas turbine engines used in the electric power industry and for gas transportation). Changing the geometry of the gas-turbine engine flow path leads to a change in the engine operating parameters. All of the above design and operational features significantly affect the qualitative and quantitative indicators of the vibration signal.

- the authors analyze the obtained cascade spectrum and identify the first rotor harmonics corresponding to a specific fault, while, as in [7], no attention is paid to the establishment and study of changes in correlations between the measurement channels in the vibration diagnostics system, which also tend to change over time and can improve the quality of diagnostic tests.

The objective of the invention is to develop a method for determining the technical state of complex technical systems, using the example of a gas turbine engine, based on statistical analysis of correlations in multichannel vibration measurement systems and analysis of higher-order spectra using the theory of invariants. In this case, the diagnosis of the object should be carried out in resource-saving modes.

The objective of the claimed invention is solved due to the fact that in the present work the possibility of using the apparatus of cumulants of higher orders to obtain a system of diagnostic signs and invariants suitable for recognizing the early stages of damage to the bearing supports of a gas turbine engine, incl. at rotor speeds corresponding to the modes of cold or technological scrolling (HP or TP) of the engine.

The mathematical apparatus for processing cumulants proposed in this invention structurally consists of two algorithms:

1) an algorithm for testing the hypotheses of linearity and Gaussianity of the process under study. In this algorithm, the hypothesis of violation of certain characteristics of linearity and Gaussianity of the process in the event of the occurrence and development of damage is realized and tested;

2) an algorithm for processing higher-order cumulant matrices using the procedure of metric scaling of diagnostic invariants in the regularized version.

To identify invariant diagnostic features when using the higher-order mathematical apparatus of spectral analysis, it is necessary to consider the general provisions of applied analysis of random data, which are required to determine the cumulants of higher-order spectra.

Let the set of sample functions x _k (t) (ensemble, where t is a variable and k is fixed) defines a random process. The average value (first moment) of this random process at time t ₁ can be calculated by taking the instantaneous values of all sample functions of the ensemble at time t ₁ , adding these values and dividing by the number

selective functions. Similarly, the covariance (mixed moment) of the values of a random process at two different times (covariance function) is calculated by averaging over the ensemble of the products of instantaneous values at times t ₁ and t ₁ + τ. Therefore, the mean value μ _x (t ₁ ) and the covariance function R _xx (t ₁ , t ₁ + τ) of the random process {x (t)} are determined by the formulas:

in which the summation is performed under the assumption that all sample functions are equally probable. In the general case, when μ _x (t ₁ ) and R _xx (t ₁ , t ₁ + τ) depend on the time t ₁ , the random process {x (t)} is called nonstationary. In the case when there is no dependence on time t ₁ , the random process is called weakly stationary or stationary in a broad sense.

The average value of a weakly stationary process is constant, and the covariance function depends only on the time shift τ, i.e. μ _x (t ₁ ) = μ _x and R _xx (t ₁ , t _{1 + τ} ) = R _xx (τ).

To determine the complete set of distribution functions that define the structure of the random process {x (t)}, it is necessary to calculate an infinite number

moments and mixed moments of higher orders. In the case when all moments and mixed moments are invariant in time, the random process {x (t)} is called strictly stationary or stationary in the narrow sense.

In most cases, the characteristics of a stationary random process can be calculated by averaging over time within the individual sample functions included in the ensemble. Take the k-th sample ensemble function. The mean value μ _х (k) and the covariance function R _xx (τ, k), calculated from the kth implementation, are equal to:

If the random process {x (t)} is stationary, and μ _х (k) and R _xx (τ, k) calculated from different realizations according to formulas (3.2) coincide, then the random process is called ergodic. For ergodic processes, the mean values and covariance functions obtained by averaging over time are equal to the analogous characteristics found by averaging over the ensemble, i.e. μ _х (k) = μ _х and R _xx (τ, k) = R _xx (τ).

For two random processes {x _k (t)} and {y _k (t)}, it is necessary to estimate their joint properties by an appropriate analysis of an arbitrary pair of sample functions x _k (t) and y _k (t).

Average values over the ensemble at an arbitrary fixed time t:

where М [] - mathematical expectation.

Correlation functions defined for arbitrary fixed times t ₁ = t and t ₂ = t _{1 + τ} :

In general, these values are different for different combinations of t ₁ and t ₂ . For τ = 0 (t = t ₁ = t ₂ ):

those. the correlation functions C _xx (t, t) and C _yy (t, t) coincide with the usual variances of the processes {x _k (t)} and {y _k (t)} at a fixed time moment t, while C _xy ( t, t) - represents the covariance of the random variables {x _k (t)} and {y _k (t)}.

For stationary random processes, you can write:

where p (x) and p (y) are the probability densities of the random variables x _k (t) and y _k (t). The correlation functions of stationary processes are independent of t.

Let us find their functions for arbitrary fixed t:

where instead of C the symbol R is used in order to distinguish these values from the correlation functions introduced by formulas (4). For nonzero means, R is different from C.

The quantities R _xx (τ) and R _yy (τ) are called autocorrelation or covariance functions of the processes {x _k (t)} and {y _k (t)}, and R _xy (τ) is the mutual covariance function {x _k (t) } and {y _k (t)}.

For two processes {x _k (t)} and {y _k (t)}, the joint probability density p (x ₁ , x ₂ ) of a pair of random variables x ₁ = x _k (t) and x ₂ = x _k (t + τ ) does not depend on t. The joint probability density p (y ₁ , y ₂ ) for a pair of random variables y ₁ = y _k (t) and y ₂ = y _k (t + τ) is also independent of t. The joint probability density p (x ₁ , y ₂ ) has the same property. In terms of these densities:

Consider the definition of spectral density. The first way is to perform a Fourier transform from a pre-computed covariance function. This approach gives a two-sided spectral density denoted by S (f) and defined for a frequency f from (-∞, ∞).

Let there exist covariance and mutual covariance functions R _xx (τ), R _yy (τ), R _xy (τ) given by formulas (7). Suppose that the integrals of their absolute values are finite:

In practice, this condition is always satisfied for realizations of finite length. Then the Fourier transform of the functions R (τ) exist and are determined by the formulas:

The quantities S _xx (f), S _yy (f) are called the spectral density functions of the processes {x _k (t)} and {y _k (t)}, and S _xy (f) is called the mutual spectral density {x _k (t)} and {y _k (t)}.

Inverse Fourier transforms from formulas (10) give:

Relations (10) and (11) are called Wiener-Khinchin formulas.

The spectral function is the initial means for processing vibroacoustic signals. The usefulness of the spectral function stems from an important theorem known as the Walsh expansion, which states that any time-discrete stationary random process can be expressed in the following form:

in which the processes у (t) and z (t) are uncorrelated and the process y (t) is a linear process:

The autocorrelation function or autocorrelation sequence of a stationary process, x _k (t), is determined by the formula (7):

The spectral function is expressed as the Fourier transform of the autocorrelation sequence:

where f is the frequency. The equivalent definition is:

where X (f) is the (Fourier) transform x (t)

A necessary, but not mandatory, condition for the existence of a spectral function is that the autocorrelation must be absolutely summable. The spectral function is real and non-negative, i.e. S _xx (f) ≥0; if x (t) is real, then the spectral function is also even S _xx (f) = S _xx (-f).

Let's introduce the definition of cumulants:

Cumulants are nonlinear combinations of higher-order moments, which, in turn, are generalizations of autocorrelation.

The first-order cumulant of a stationary process is the average value

Higher-order cumulants are invariant with respect to the shift of the mean value; therefore, it seems convenient to define them under the condition of a zero mean.

Cumulants of the second, third and fourth order of the zero mean of a stationary process are determined by the formulas:

where E _2x (m) = M [x (t) x (t + m)], and is equal to C _2x (m), for a real process. Thus, the first order cumulant is the process mean; and the second-order cumulant is an autocovariance sequence.

Let us introduce special designations for the zero-delay cumulants:

- C _2x (0) - represents a variance and is usually denoted as σ ² _x ;

- C _3x (0,0) and C _4x (0,0,0) are usually designated as γ _3x and γ _4x .

Let us turn to the normalized values, γ _3x / σ ³ _2x - asymmetries and γ _4x / σ ⁴ _2x - excess. These both normalized quantities are offset and scale invariant. If x (t) is symmetrically distributed, then its asymmetry is necessarily equal to zero (but not vice versa); if x (t) is normally distributed, then its kurtosis is necessarily zero (but not vice versa). Often the terms "skewness" and "kurtosis" are used to denote the quantities γ _3x and γ _4x .

If x (t) is statistically independent with respect to y (t), then

characterized by the same dependencies that remain for the cumulants of all orders.

This additivity property greatly simplifies the cumulant-based analysis.

The cumulants of a stationary real process are symmetric with respect to their arguments, that is:

The k-order poly-spectrum is defined as the Fourier transform of the corresponding sequence of cumulants:

which respectively represent spectral function, bispectrum and trispectrum. Note that bispectrum is a function of two probability densities, while trispectrum is a function of three densities. Unlike the spectral function, which is real and non-negative, bispectra and trispectra are complex-valued.

For a real process, a transition from the symmetry of the cumulant to the symmetry of polyspectra is characteristic. The spectral function is symmetric (multiple): S _2x (f) = S _2x (-f). The bispectrum symmetry is:

, выражена как (m_k=(m₁…,m_k-1)):The reason for the use of cumulants and polyspectra of order

, expressed as (m _k = (m ₁ ..., m _k-1 )):

If z (t) = x (t) + y (t), ax (t) and y (t) are independent processes, then C _kz (m _k ) = C _kz (m _k ) + C _ky (m _k ) ...

.If x (t) has a one-dimensional Gaussian distribution, then C _kz (m _k ) = 0,

...

, C_kz(m_k)=C_kx(m_k). Таким образом, возможно восстановить кумулянты высшего порядка негауссовского сигнала даже в присутствии гауссова шума.Therefore, if z (t) = x (t) + w (t), where w (t) is normal and independent of x (t), then, for

, C _kz (m _k ) = C _kx (m _k ). Thus, it is possible to recover the higher order cumulants of a non-Gaussian signal even in the presence of Gaussian noise.

Let x (t) be a linear process, that is, x (t) = Σ _τ h (τ) u (t-τ), then it follows that:

where γ _ku = C _ku (0). Note that the spectral function does not carry any information about the phase of H (f). In contrast, if the process u (t) is not Gaussian, then this information can be retrieved from higher order polyspectra.

Any process can be considered a linear process with respect to its second-order statistics; that is, for a given R _yy , one can always find {h (τ)} and an uncorrelated process u (t) such that R _{yy (} m) = R _xx (m), where x (t) = Σ _k h (τ ) u (t-τ). In other words, it is not possible to determine the characteristics of nonlinearity from the autocorrelation sequence. In contrast, higher-order cumulants allow this to be done.

Since the second-order cumulant C _2x (τ) is the correlation function of the process x (t), the second-order polyspectrum is the usual power spectrum. Similarly, the third-order cumulant coincides with the third moment relative to the mean.

The main property of polyspectra is that if the process x (t) is Gaussian, then all its polyspectra, starting from the third order, are identically zero, so they can be used as a measure of the deviation of the finite-dimensional distributions of the process from Gaussianity.

Thus, if a process has a nonzero bispectrum, this may be due to the following reasons:

- the process obeys a linear model with non-Gaussian shaping noise;

- the process obeys a non-linear model, and the noise can be both Gaussian and non-Gaussian.

The bispectral function has the properties of suppressing noise components, since the third-order moment of the Gaussian process is equal to zero, at the same time it reacts to the dependence between discrete frequency components f ₁ , f ₂ , f ₃ , if they satisfy the relation f ₁ + f ₂ + f ₃ = 0, where f ₃ = - (f ₁ + f ₂ ).

The value of the information obtained from the bispectrum lies in the fact that this characteristic helps to understand the structure of the oscillatory process, to detect parts of the spectrum that are statistically related to each other, and also to reveal the presence of combination and modulation frequencies.

In contrast to the power spectrum, a real positive function of frequency, the bispectrum is a complex quantity, therefore, the bispectrum module is most often used for diagnostic purposes, which can be represented as a generalized diagnostic feature with its representation in matrix or vector form with a sequential scan along columns and rows.

With a discrete nature of oscillations, the bispectrum is a highly sensitive characteristic of changes in the parameters of the technical state of rotor mechanisms.

It is not uncommon for a change in the parameter of the technical state of the mechanism to cause small energy changes in the acoustic signal, at the same time affecting the phase relationships between multiple frequency components of the signal. It is impossible to capture these changes in the spectrum, while the bispectrum allows to reveal phase information even in the presence of noise interference.

The considered properties of polyspectra, incl. bispectra, allow to implement two algorithms for assessing the technical condition of GTE supports on resource-saving modes of operation:

1) Algorithm for testing the hypotheses of linearity and Gaussianity of the process under study;

2) Algorithm for processing higher-order cumulant matrices using the procedure of metric scaling of diagnostic invariants in the regularized version.

For an understanding of the theoretical part of the proposed invention, consider examples of the implementation of this method.

Example 1. To identify diagnostic invariants when using a higher-order spectral analysis apparatus, a computational experiment was carried out on the mathematical model of the vibration process of the GTE rolling bearings based on the following considerations:

Consider a set of sample functions x _i (k) (t) _{in the} form of simulated time series for 5 damages:

- violation of the geometry of the raceways;

- rolling element defect;

- the separator touches the inner ring;

- defect in the outer ring;

- separator defect. We will consider the process to be stationary. For each i-th damage, the modeled (observed) time series in the form of a column matrix x _i (k) are distorted by noise, so that the signal-to-noise ratio P signal _. / R _noise = h varied randomly within [h _min ; h _max ].

In this case, the model also plays out the amplitudes of defects, the rotational speed of the rotors corresponding to the values of the frequencies of cold or technological scrolling (from 900 to 1800 rpm). The number of implementations for the considered examples was limited to 50.

FIG. 1 and FIG. 2 shows examples of temporal realizations and frequency spectra for two damages with different levels of simulated noise (where: a) and b) - temporal realizations at different levels of induced interference; c) the frequency spectrum of the implementation (b)).

As a result of one-factor experiments, sets of implementations were obtained in the form of the following array:

For each realization x _i (k) with a length of 1024 values, cumulants of the 3rd order C _3x = cum _{3 (i)} (x _{i (k)} ) are calculated on a window with a width of 1 equal to 8,16,32,64,128 values. In general, the length of the row can be any, but it is desirable that the length of the input data sequence be a power of 2 (1024,2048,4096, etc.), since in this case, the cumulant calculation algorithm has the maximum time performance (by analogy with the fast Fourier transform procedure) (for 1024 points - 11 s, for 1000 points - 130 s)

Calculations have shown that the optimal window width for arrays of the indicated dimension is l = 32.64 values. The obtained values of the cumulants cum _{3 (i)} (x _{i (k)} ) for each row of the array X _{i (k) are} written in the form of a training matrix of the corresponding damage (for 1024 values of the original array, the dimension of the cumulant matrix is 33 × 50):

The calculated matrices of the third-order cumulants are “reference” portraits of damage, which are subsequently used to implement the multidimensional scaling procedure in a regularized version, similar to diagnostics by thermogasdynamic parameters. After calculating the corresponding matrices cum _{3 (i)} (x _{i (k)} ) according to the vibration process model for 5 bearing damages, hodographs of single defects were constructed (Figs. 3, 4 and 5). Thus, in FIG. 3 and 5 the following types of defects are presented under digital designations: 1 - violation of the geometry of the raceways; 2 - defect of the rolling body; 3 - the separator touches the inner ring; 4 - defect of the outer ring; 5 - separator defect.

The calculation of cumulants and the construction of hodographs of damage to the bearing units of the gas turbine engine were carried out using the MATLAB software environment.

The search for the optimal projection of the hodographs as a result of the rotation operation in three-dimensional space makes it possible to find (calculate) the image with the best resolving properties. It should be taken into account that the maximum peaks (extreme areas of hodographs) correspond to the maximum amplitudes of the modeled process, while the signal-to-noise ratio _Psign. / P _noise = h determines the width of the hodograph and the degree of discernibility of defects.

Taking into account the fact that during the simulation, a noise level was specified that distorts a signal with a lower h than in a real experiment, it can be assumed about the corresponding invariance of the hodograph beam to real noise, which will be shown below.

It should be noted that in the proposed method for assessing the technical condition of GTE supports, noises are not considered as diagnostic signs due to the fact that this method assumes the maximum elimination of various kinds of interference when diagnosing GTE in resource-saving diagnostic modes (HP, rotor run-out mode).

FIG. 6 and 7 show the results of visualization of the hypotheses of linearity and Gaussianity of the studied processes for one and several realizations obtained on the mathematical model of the vibration process. With reference numerals in FIG. 7 shows the following types of defects: 3 - contact with the separator of the inner ring; 5 - separator defect.

In the presence of data obtained in the course of long-term measurement of vibroacoustic characteristics (i.e., in the presence of archived data), it is possible to perform a complex diagnostics procedure based on the calculation of the corresponding invariants and the construction of the spatial structure of damage hodographs.

The above diagnostic methods and algorithms for their implementation made it possible to verify the possibility of their practical application based on the results of mathematical modeling of the processes occurring in a gas turbine engine.

Example 2. Also, the developed methods were tested and their performance checked on full-scale objects.

To implement the proposed method at the objects of diagnostics, it is important to measure vibration in the same places, called control (standard) vibration measurement points (usually on the bearing supports of the unit, the unit casing and on the anchor foundation bolts). For each gas turbine unit, the control points are determined guiding documents (technical descriptions, operating rules, etc.).

It is recommended to measure absolute vibration (when diagnosing most mechanical defects) in three mutually perpendicular directions: vertical, horizontal - transverse and axial. The transducers for measuring the horizontal - transverse component of vibration are fixed at the level of the shaft axis against the middle of the length of the support insert. The axial component of vibration should be measured at a point as close as possible to the shaft axis on the bearing support housing near the horizontal joint between the cover and the housing. The vertical component of vibration is measured at the top of the bearing cover above the center of the bearing shell.

It is allowed to measure the vertical, horizontal and axial components of vibration by installing a three-component vibration sensor on the upper part of the bearing cover for measuring vibration in mutually perpendicular directions coinciding with the main axes of the unit.

If it is impossible to carry out measurements in three main directions in the area of one bearing or it is required to minimize the number of measurements, then vibration measurement in two directions is allowed: axial and one of the transverse directions. The preference is given to the transverse direction, as a rule, corresponding to the direction of the minimum rigidity of the system. It is also allowed to measure axial vibration of the drive, blower and other units of the unit only at the bearing of the free shaft end.

The gas turbine engine DR-59L No. DOO299456 gas pumping station "Pikalevo" was chosen as the object of diagnostics. It should be noted that for this engine, a complete set of dispatch and operational documentation was available, according to which it was possible to trace all stages of the product life cycle. The general characteristics of the DR-59L engine No. D00299456 are presented in tabular form (Fig. 8).

The inspection of the gas turbine engine at the gas pumping station in Pikalyovo was carried out from February to May 2016, followed by disassembly and inspection of the engine at the Kronstadt Marine Plant in July-September 2016.

In agreement with the management of Gazprom Transgaz Saint Petersburg, an instrumental examination of the gas turbine engine was carried out. The engine inspection program included an assessment of its technical condition at the CP using the methods developed in this work.

Diagnostic procedures were instrumented by bringing an acoustic sensor through the oil drain from the GTE support using the hardware of Gazprom Transgaz Saint Petersburg and Orgtekhdiagnostika LLC.

The repeated instrumental examination was carried out using the standard equipment of Gazprom Transgaz Saint Petersburg.

FIG. 9 shows the result of comparing the experimental damage travel time curve with travel time curves obtained on the basis of calculations based on the geometric characteristics of the inter-shaft bearing.

The research carried out by instrumental methods confirmed the preliminary diagnosis of the destruction of the bearing of the intermediate support of the gas turbine engine. In the joint act of technical examination of the engine, it was recommended to stop the operation of the engine. GTE DR-59L head. No. DOO299456 was dismantled and sent to KMOLZ (Kronstadt). As a result of disassembly and fault detection of the engine, it was established:

1) destruction of the inner ring and bearing cage of the intermediate support;

2) development of the inner surface of the splined sleeve;

3) destruction of labyrinth seals on the outer surface of the sleeve.

Thus, the results of flaw detection fully confirm the preliminary diagnosis, and timely dismantling and sending the engine to a repair facility prevented more serious damage.

The method of diagnostics proposed in the work, as well as its software, algorithmic and hardware support are used:

1) in the formation of a unified system for technical diagnostics of gas turbine engines as part of gas-pumping units of OOO Gazprom Transgaz Saint Petersburg

2) in the "Program of technical diagnostics and inspection of gas turbine installations of surface ships of the Navy" NPO Saturn

The essential features of the invention are:

a) diagnostics of the gas turbine engine both in working and resource-saving modes of operation;

b) the possibility of recognizing (revealing) the moment of initiation of a failure at an early stage, when there were no obvious tendencies in the change of parameters and disturbances in the operation of the gas turbine engine;

c) determination of changes in correlations between measurement channels and establishment of a preliminary diagnosis of the object to be diagnosed;

d) confirmation of the diagnosis by instrumental methods of diagnosis and defect detection at repair enterprises.

Thus, the proposed method can be used to monitor the vibration state of gas turbine engines of sea and river transport, ships and vessels of the Navy, and can also be used for vibration diagnostics of engines used in the electric power industry and for gas transportation.

Literature:

1. USSR author's certificate No. 1107002, MKI G01H 1/00. A device for vibroacoustic diagnostics of intermittent mechanisms. / Kostyukov V.N. and Morozov S.A. // App. 04.17.80; Publ. 08/07/84; Bul. - No. 29.

2. USSR author's certificate No. 1000776 MKI G01H 13/00 Method of vibration diagnostics of turbomachines. / Doroshenko S.M., Chemokhud E.V. /. Appl. 05.20.80; Publ. 02/28/83; Bul. - No. 8.

3. USSR author's certificate No. 1652861 MKI G01M 15/00 Turbomachines diagnostics device. / Ryndak V.K., Sidorin G.N. /. Appl. 05/16/89; Publ. 05/30/91; Bul. - No. 20.

4. Biryukov R.V., Kisilev Yu.V. Vibration diagnostics of rotary bearings of gas turbine engines using cold cranking modes. - M .: Mechanical engineering. /https://cyberleninka.ru/viewer_images/16908885/f/4.png/

5. Bogdanov S.S., Kolesnik V.A., Makshanov A.V. Detection and classification of defective states of machines and mechanisms based on the results of multichannel measurements of various physical nature. - "Ecology and nuclear energy". Collection of works of MANEB, St. Petersburg: 1999, p. 37.

6. RF patent No. 1343259, MKI G01M 7/00. A device for vibroacoustic diagnostics of intermittent mechanisms. / Kostyukov V.N. / Appl. 02.24.86; Publ. 10/07/87 .; Bul. - No. 37.

7. RF patent No. 2337341, MKI G01M 15/00 (2006.01), G01M 7/02 (2006.01). Method of vibrodiagnostics of the technical condition of piston machines by spectral invariants / Kostyukov V.N., Naumenko A.P., Boychenko S.N. / Appl. 04/11/2007; Publ. October 27, 2008.

Claims (1)

The method of vibrodiagnostics of the technical state of gas turbine engines in resource-saving modes using the theory of invariants is based on the measurement and analysis of information arrays of vibroacoustic parameters of the diagnostic object, obtained in the modes of cold or technological engine scrolling, establishing and studying the change in correlations between the measurement channels in the vibration diagnostics system using the theory of invariants and cumulants of spectra of higher orders, and as a result - determination of the technical state of the object in the course of comparing the research results with their reference values.

Images